GRADUATE STUDENT SEMINAR, SPIDER: Sparse Physics-Informed Discovery of Empirical Relations, Daniel Gurevich

SPIDER: sparse physics-informed discovery of empirical relations

In recent years, there has been a surge of interest in interpretable regression-based machine learning of physics models from data, as popularized by the SINDy algorithm by Brunton et al. However, such methods have been largely limited by conceptual shortcomings in tackling real-world physical and biological systems. In this talk, we describe SPIDER, a general algorithm for model discovery that resolves many of these limitations, allowing much broader success. Our approach unites techniques from programming language theory, PDEs, discrete optimization, and mean-field theory and is illustrated using a range of examples from hydrodynamics.